When off-the-shelf AI can't follow your rules, I build what can.
Custom AI and software development for small and mid-sized companies: internal tools, portals, AI agents and full platforms, built around how your business works, with checks and a person approving.
Custom AI development means building software around your rules when off-the-shelf tools can't follow them. I build internal tools, portals, AI agents that hand decisions to a person, connections between AI and your existing systems, and full platforms. Larger builds are phased (prototype, pilot, production), each with a fixed fee quoted in writing, so you can stop at any decision point. Every build has checks: rules for each output, automated tests, and my own review before release. Examples include a live AI-assisted broker portal for a yacht brokerage, and a healthcare platform with 500+ automated tests (my own venture, in development).
- Find it
- Quick win
- One workflow, end to end
- Connected systems
- Custom AI tool or agent
- Platform or product build
- Run and improve
When custom is worth it
Start with what you already pay for. ChatGPT, Claude or a no-code automation often handles the simple part of the work. Custom AI solutions for small businesses are worth it when one or more of these is true:
- Your rules are unusual: approvals, pricing logic, or what may be published.
- The work crosses several systems or several companies.
- One source has to feed many outputs that must all match, like languages, documents or channels.
- Different people must see different things, so roles and permissions matter.
- The off-the-shelf version needs so much copying and checking that it saves nothing.
If a tool you already have can do it, I'll say so. Sometimes the fix is a better process.
If you're not sure custom is needed, start with an opportunity review. How we work
If it's one task, start smaller: one task, working in about four weeks
What I build
I build and ship software with AI. This is bespoke software development for small and mid-sized companies, built around your rules and your people.
Internal tools and portals
A purpose-built place where your team does one job properly, with structured records, drafts and approvals, and a clear status for every item.
I built the website and broker portal for a yacht brokerage. In the portal, the assistant, Athena, proposes listing fields and shows the source wording beside each one. Unsupported numbers and unclear currencies or units are held for the broker to check.
For example (illustrative): a quoting tool that follows your price rules, or an order-review screen for your team.
AI agents that hand decisions to a person
An agent can prepare a defined action, such as changing a price, starting a workflow or drafting a reply. A person confirms it. The confirmation is tied to that record and rechecks its current value first. The agent can't publish, delete or make unrestricted bulk changes on its own.
This is how I approach AI agent development for a business: each agent prepares the work, and a person makes the call.
Connecting AI to your data and systems
AI that reads from and writes to the systems you already have: databases, CRM, accounting, document stores and websites. It starts read-only, and write access comes only with your approval. This is how I integrate AI with existing systems, older ones included.
Where a system has no modern API, we agree a safe route (an export, a scheduled file, a controlled screen step) before building anything.
For simpler connections between your systems, see workflow automation and integration
Full platforms and products
Applications for several roles and several languages, with permissions, a fixed set of stages and an automated test suite.
The healthcare platform is my own venture, in development. It has 4 workspaces (patient, prescriber, clinic and pharmacy), 3 languages (English, Italian and Greek) and 500+ automated tests, including 15 that walk a full journey role by role. Access follows the relationship, and every next step has an owner.
Build and test inventory, September 2026.
Built around the way you work
We start with what needs to improve, not with the technology.
I look at who touches the work, what they decide and what goes wrong. Your documents, rules and approvals become the software's rules. Before anything goes live, your people test it on real and awkward examples, and we fix what they find.
AI can prepare the content. The rules decide which text it is allowed to touch.
Case study: an AI-assisted broker portal for a yacht brokerage
The hard part for a yacht brokerage is keeping hundreds of generated pages, images and PDFs correct, week after week.
I built the brokerage's website and broker portal around one structured record for each yacht, with a built-in assistant, Athena. Approved content goes into pages in English, Greek and Russian, a public photo set, a branded brochure for clients and an unbranded brochure for other brokers.
80+ yachts · 240+ listing pages · English, Greek, Russian · 2 brochures per yacht
Live sitemap count, 24 September 2026
The publishing workflow is live.
"Saved", "approved" and "on the website" never mean the same thing.
Read the full case study“I don't know very much about technology, but I can certainly see the excellent result, and the care you've put into helping [the business] grow.”
Founder and broker, boutique East Mediterranean yacht brokerage · translated from Greek
Case study: a healthcare platform
My role: Co-founder. I lead the product and direct the build. A licensed pharmacist co-founder keeps it grounded in how pharmacies really work.
Patients, prescribers, clinics and pharmacies each have their own workspace and see only what they should. A request moves through fixed stages: review, quote, accept, reject, cancel or expire. It's routed by country and by what each pharmacy can actually supply.
More than 500 automated tests check the build, including 15 that walk a whole journey the way a patient, doctor, clinic or pharmacy would. Tests show what was checked, not clinical outcomes. Clinical and dispensing decisions always stay with licensed professionals.
Build and test inventory, September 2026.
If your work passes between several people (orders, claims, approvals), the same approach applies.
Read the case studyNeed something like this built?
Who builds it
I build and ship software with AI. On the healthcare platform I designed the workflows, directed AI to write and check the software, and tested each role's journey with made-up patients, doctors and pharmacists before a real one came near it.
For you, that means one accountable person who understands both the business and the build. The person who plans the work is the person who builds it, so nothing gets lost in a hand-off.
I've spent 27 years on the business side of technology, including Director of IT at Warner Music Group in Los Angeles. Larger builds are phased, and I lead and build them.
About MichaelHow a build runs: prototype, pilot, production
This is how a build moves from proof of concept to production, one phase at a time.
Prototype
We prove the hard part on your real examples. Then you get a written decision: go, change or stop.
Pilot
Real users do real work on a limited scope, and a person approves every output. We measure time per task before and after, including checking and corrections. Those figures are yours, and I don't publish them.
Production
Release, a named owner, written instructions and handover training. Support is as agreed in the engagement.
Each phase is quoted in writing as a fixed fee before it starts, and you can stop after any phase. Timing is agreed in writing for each phase, once we've checked what the work depends on.
Checks I build in
- Rules for each output: what AI may touch, and what it may not.
- Holds instead of guesses. A missing source, conflicting units or a record that changed during review stops the process until a person looks. Nothing is made up, and nothing is published without a person knowing.
- Automated tests before each release, including tests that walk whole journeys.
- Drafts kept separate from what's live.
- Automatic checks, then my own eyes.
- A person approves anything that leaves the business or changes a record.
What you get, and after launch
- Working software in use, with a named owner, written instructions and handover training.
- You own the code, workflows and data built for your business. Account ownership, licences and support are set out in the proposal.
- The AI providers I use are OpenAI, Anthropic (Claude) or xAI (Grok), chosen for the task. Running costs, hosting and account ownership are stated in the proposal.
- Support after launch is as agreed in the engagement. Further development can continue under a fractional Head of AI retainer.
Questions
When is custom AI development worth it?
When your rules, approvals or outputs are unusual, when work crosses several systems, or when off-the-shelf tools need so much copying and checking that they save nothing. If a tool you already have can do the job, I'll say so.
What can you build?
Internal tools and portals, AI agents that prepare actions for a person to confirm, connections between AI and your existing systems, and full multi-role platforms with automated tests. Each one is built around your rules, and a person approves anything that changes a record or leaves the business.
How do you keep AI output correct?
Each output has its own rules. Missing or conflicting information is held for a person, not guessed. Automated tests run before every release, and I check the result myself. A person approves anything that changes a record or leaves the business.
Can you work with our existing or older systems?
Usually, yes. We start read-only, agree a safe route for each system and add write access only with approval. Where a system has no modern API, we agree the route before building anything, such as an export, a scheduled file or a controlled screen step.
How long does a custom build take?
It depends on the work. Larger builds are phased (prototype, pilot, production), and timing is agreed in writing for each phase once we've checked what the work depends on. Each phase ends with a written decision, so you can stop after any phase.
How much does custom AI development cost?
Each phase is a fixed fee, quoted in writing before it starts. What drives the fee: the number of systems, the state of your data, the approval steps, the languages, the testing and the support. Running costs are stated in the proposal.
Who owns the code?
You own the code, workflows and data built for your business. Account ownership, licences and support are set out in the proposal, so it's clear before we start which accounts and third-party software are yours and who looks after what.
Which AI models do you use, and where does our data go?
I work with OpenAI, Anthropic (Claude) and xAI (Grok), all US providers, chosen for the task. We agree up front what data the work needs, what the AI may see and do, and where a person must decide. Ask me how your data would be handled for your build.
What happens after launch?
A named owner in your team runs it, with written instructions and handover training, so your people can use it without me. Support after launch is as agreed in the engagement, and further development can continue under a fractional Head of AI arrangement.
How can one person build all this?
I design the workflows, direct AI to write and check the software, and test every journey before real users see it. You get one accountable person, and nothing is handed over between the person who plans the work and the person who builds it.
Got a process off-the-shelf tools can't handle? Bring it to a 15-minute call.
Tell me one task your team hates doing.
In a free 15-minute call, we'll work out whether AI can take it on and what a sensible first step would be. If the answer is no, I'll say so.
Free · 15 minutes · you'll talk to me, not a sales team.
